multi-rgbd-inertial-dataset

KO | EN

gitlite — search

#python #java #python3 #arduino #golang #machine-learning #rust #html #flask #javascript #seismology #nodejs

multi-rgbd-inertial-dataset

★ 37 Open GitHub ↗

Dataset for RA-L 2025 paper “The More The Better? Confidence-Driven Residual Weighting and Depth Fusion for Multi-RGB-D Inertial Odometry” - synchronized 4-camera RGB-D + IMU sequences.

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Multi-RGB-D Cameras and Inertial Dataset (with Hyundai Motor Company)

**A comprehensive collection of time-synchronized RGB-D with IMU and LiDAR recordings captured by four Intel RealSense D455 cameras, a MicroStrain GX5-25 IMU, and an Ouster-64 LiDAR. Covering 360° non-overlapping views and diverse indoor/outdoor scenarios, this dataset enables research on confidence-driven multi-camera odometry, depth fusion, and scale-consistent VIO/SLAM.**

<sup>Created in collaboration with Hyundai Motor Company and released alongside our RA-L 2025 paper below.</sup>

<h2>The More The Better? Confidence-Driven Residual Weighting and Depth Fusion for Multi-RGB-D Inertial Odometry</h2>

[**Seungsang Yun**](#) · [**Jaeho Shin**](#) · [**Jaekwang Cha**](#) · [**Ayoung Kim**](#*)

<sup>Seoul National University / Hyundai Motor Company</sup>

**RA-L 2025 (to appear at ICRA 2026)**

[[`Paper PDF`](https://ieeexplore.ieee.org/abstract/document/11130931)] [[`Project Video`](https://www.youtube.com/watch?v=eTlhmUA6c64)] [[`BibTeX`](#citation)]

</div>

---

## 1. Dataset Features
* **Synchronized multi-sensor suite** – four hardware-triggered Intel RealSense D455 RGB-D cameras and a IMU
* **360° non-overlapping coverage** – front, back, left, and right views enable true multi-camera VO/VIO/SLAM research  
* **Challenging scenarios** – sequences include strong glare, low-light, texture-less walls, and dynamic occlusion in both indoor and outdoor environments  
* **Complete calibration package** – intrinsics, extrinsics, and camera-to-IMU SE3 transforms provided  
* **Research-ready formats** – ROS bag files, ground truth

## 2. Sensor Suite

| Component | Model | Rate | Notes |
|-----------|-------|------|-------|
| **RGB-D cameras (×4)** | Intel RealSense D455 | 640 × 360 @ 30 Hz | Front / Left / Right / Back (non-overlapping FOV) |
| **IMU** | MicroStrain 3DM-GX5-25 | 100 Hz |  |
| **LiDAR** | Ouster-64 | 10 Hz | Provides ground-truth via LiDAR-Inertial SLAM |

---

## 3. Sensor Synchronization

<p align="center">
  <img src="figures/rqt.png" width="90%" alt="ROS rqt_plot showing camera sync">
</p>
<p align="center"><sub><em>rqt_plot screenshot confirming sub-millisecond timestamp alignment across the four RGB-D camera topics.</em></sub></p>

All images and depth frames are hardware-triggered.

## 4. Sequences(The link will be accessible shortly.)
| ID         | Length        | Challenges (✓)                               | Dataset |
|------------|---------------|----------------------------------------------|---------|
| `indoor`   | 100 m / 104 s | corridor, moderate texture                  | [link](https://drive.google.com/file/d/1uy7-UXpa3LDM8zhQ6uEZnleFtaGFYbV7/view?usp=drive_link) |
| `indoor2`  |  69 m / 159 s | spacious indoor, occlusion ✓                | [link](https://drive.google.com/file/d/1cwh9N4ehBy2w2u-9lLux0SMfGlBe3Zy7/view?usp=drive_link) |
| `glare`    |  45 m / 155 s | strong sunlight glare ✓                     | [link](https://drive.google.com/file/d/1bBzP2LWhxftfbWACPSUxVSKGy0pAQTau/view?usp=drive_link) |
| `txl`      |  68 m / 150 s | texture-less walls ✓ glare ✓ dark ✓         | [link](https://drive.google.com/file/d/1DsSfANcHsrjoOXbacNhn3LNscDgQmqW_/view?usp=drive_link) |
| `occ`      |  67 m / 146 s | front-camera occlusion ✓                    | [link](https://drive.google.com/file/d/1VnVaayk-2bhdT6Nzwy5gKmpAONFdHDQX/view?usp=drive_link) |
| `dark`     |  73 m / 169 s | low light, large rotations ✓ texture-less ✓ | [link](https://drive.google.com/file/d/1p01MbeDzZnwGI_EfiTz5jC8zHcqiqZT1/view?usp=drive_link) |
| **Calibration**  | –             | intrinsics & extrinsics               | [link](https://drive.google.com/file/d/1-lHyAXZdq-Ik-u_toySdV52fC5onzmk7/view?usp=drive_link) |
| **Ground truth**| –             | LiDAR-IMU SLAM trajectories (tum foramt)| [link](https://drive.google.com/file/d/1K9YL5tfDt0aUgRUCS5C_iGVwlo5rp0ag/view?usp=drive_link) |

---
## 5. Example of Multi-camera images

#### Sequence. Indoor #1
![Indoor #1 – LiDAR projected](https://raw.githubusercontent.com/seungsang07/multi-rgbd-inertial-dataset/main/figures/indr.png)

#### Sequence. Indoor #2
![Indoor #2 – LiDAR projected](https://raw.githubusercontent.com/seungsang07/multi-rgbd-inertial-dataset/main/figures/indr2.png)

#### Sequence. Glare
![Glare scene – LiDAR projected](https://raw.githubusercontent.com/seungsang07/multi-rgbd-inertial-dataset/main/figures/glare.png)

#### Sequence. Texture-less
![Texture-less scene – LiDAR projected](https://raw.githubusercontent.com/seungsang07/multi-rgbd-inertial-dataset/main/figures/txl.png)

#### Sequence. Occlusion
![Occlusion scene – LiDAR projected](https://raw.githubusercontent.com/seungsang07/multi-rgbd-inertial-dataset/main/figures/occ.png)

#### Sequence. Dark
![Low-light scene – LiDAR projected](https://raw.githubusercontent.com/seungsang07/multi-rgbd-inertial-dataset/main/figures/dark.png)

## 6. Multi-cameras and LiDAR Calibration examples

![Indoor #2 – projected cloud](https://raw.githubusercontent.com/seungsang07/multi-rgbd-inertial-dataset/main/figures/cam_lidar/indr2.png)
![Indoor #1 – projected cloud](https://raw.githubusercontent.com/seungsang07/multi-rgbd-inertial-dataset/main/figures/cam_lidar/indr.png)
![Glare scene – projected cloud](https://raw.githubusercontent.com/seungsang07/multi-rgbd-inertial-dataset/main/figures/cam_lidar/glare.png)

## 7. Citation

If you use this dataset, please cite our RA-L 2025 paper:

```bibtex
@ARTICLE{11130931,
  author={Yun, Seungsang and Shin, Jaeho and Cha, Jaekwang and Kim, Ayoung},
  journal={IEEE Robotics and Automation Letters}, 
  title={The More the Better? Confidence-Driven Residual Weighting and Depth Fusion for Multi-RGB-D Inertial Odometry}, 
  year={2025},
  volume={10},
  number={10},
  pages={10402-10409},
  keywords={Cameras;Odometry;Real-time systems;Optimization;Tracking;Filtering;Vectors;Robot vision systems;Pose estimation;Pipelines;RGB-D Inertial odometry;multi-camera SLAM},
  doi={10.1109/LRA.2025.3601033}
  }

```

## 8. Acknowledgement
We are deeply grateful to Hyundai Motor Company for their close collaboration and generous support throughout this project.

## 9. Contact
For any questions or issues, please contact **Seungsang Yun** at <seungsang@snu.ac.kr>.

multi-rgbd-inertial-dataset

Explore Similar Repositories

Was this content helpful?

Recent Feedback

Download README